Dataset for the analysis is obtained by FINA (Croatian Financial Agency) which collects company JOPPD reports with a number of standardized data. Since Croatian companies are obliged to consign these reports annually under threat of penalty, the dataset used here covers the entire statistical population.

The report outline changed several times. Therefore an adjustment between certain years was needed. After the adjustment, a 301 variable dataset is obtained in the time period from 2008 – 2014 for 1007 legal entities which produced non-perennial agricultural products. The unbalanced panel data set was used to estimate a Cobb-Douglas production function, which is the most commonly used in similar analyses.

Armagan & Ozden (2007) analyzed Turkish agriculture, namely crops, using a Cobb-Douglas function to estimate its production function. They used a number inputs in that study among which are the average age of farmers, their average education and land size and distinguished small, medium and large producers. The analysis was based on cross section data.

Echevarria (1998) constructed a production function for Canadian agriculture. In this paper a very common assumption was taken: scale elasticity ε = 1 (constant returns to scale) and that production elasticities of each input correspond to its share in total costs.

Parlinska & Dareev (2011) estimated agricultural production function for Poland and Republic of Buryatia. A simple two-input Cobb-Douglas function was used to estimate production functions for both countries/regions using a time series from 2000 – 2009.

Enaami, Mohamed and Ghani (2013) have shown even more advantages of using Cobb-Douglas function as a basis for production function estimation. They also show how to deal with multiple issues that might occur under a multiple input approach. Due to these suggestions, a following simple model is used to estimate Croatian non-perennial production function:

Where x stand for the legal entity (company), t for year, Y for production volume, A for total factor productivity, K for capital, L for labour, κ for contribution of capital (production elasticity of capital) and λ for contribution of labour (production elasticity of labour). Also, it is assumed that total factor productivity changes in time with an exponential time path:

Combining (1) and (2) the following function is estimated:

After linearization the estimated model was:

A generalized least squares method was used with random effects, due to abundant dataset and expected differences between the companies. Multicollinearity, heteroscedasticity and autocorrelation test were made as well as the parameter and joint tests for validity of the model.

Using the obtained data total factor productivity is calculated as a residual:

This was the end of the first stage. In the second stage a total factor productivity model was constructed using numerous regressors:

Among many, the following regressors were taken into account: share of company on the market, number of companies on the market, export volume, subsidies taken (Kroupová & Malý (2010) show the importance of subsidies on Czech agriculture, again using a multiple input Cobb-Douglas production function), growth of the economy, investment volume and many others.